A method to transfer artificial intelligence-based control from simulation to real world
Mohammadi, Ali (2021)
Diplomityö
2021
School of Energy Systems, Konetekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2021120759340
https://urn.fi/URN:NBN:fi-fe2021120759340
Tiivistelmä
Transferring the simulation model to the real world is a big challenge, since the simulation work and real-world implementation are not matched to gather usually, and it is happening due to some inaccuracies and neglection of some important parameters in the real and simulation model. Therefore, finding a suitable method to transfer the simulation model to practical work is necessary.
This dissertation work, concerned with finding a method to transfer artificial intelligence-based control from the simulation to the real world. The simulation model in this work is a four-legged robot and this robot is the physical version of the Ant benchmark of the OpenAI gym library. Firstly, this robot was trained in the MuJoCo environment and then trained in the real world, finally, both results of simulation and practical implementation were compared together to find the optimal method.
This dissertation work, concerned with finding a method to transfer artificial intelligence-based control from the simulation to the real world. The simulation model in this work is a four-legged robot and this robot is the physical version of the Ant benchmark of the OpenAI gym library. Firstly, this robot was trained in the MuJoCo environment and then trained in the real world, finally, both results of simulation and practical implementation were compared together to find the optimal method.